Implementing machine learning techniques for continuous emotion prediction from uniformly segmented voice recordings.

Introduction: Emotional recognition from audio recordings is a rapidly advancing field, with significant implications for artificial intelligence and human-computer interaction. This study introduces a novel method for detecting emotions from short, 1.5 s audio samples, aiming to improve accuracy and efficiency in emotion recognition technologies. Methods: We utilized 1,510 unique audio samples from two databases in German and English to train our models. We extracted various features for emotion prediction, employing Deep Neural Networks (DNN) for general feature analysis, Convolutional Neural Networks (CNN) for spectrogram analysis, and a hybrid model combining both approaches (C-DNN). The study addressed challenges associated with dataset heterogeneity, language differences, and the complexities of audio sample trimming. Results: Our models demonstrated accuracy significantly surpassing random guessing, aligning closely with human evaluative benchmarks. This indicates the effectiveness of our approach in recognizing emotional states from brief audio clips. Discussion: Despite the challenges of integrating diverse datasets and managing short audio samples, our findings suggest considerable potential for this methodology in real-time emotion detection from continuous speech. This could contribute to improving the emotional intelligence of AI and its applications in various areas.

Products of Variables in Structural Equation Models.

A general method is introduced in which variables that are products of other variables in the context of a structural equation model (SEM) can be decomposed into the sources of variance due to the multiplicands. The result is a new category of SEM which we call a Products of Variables Model (PoV). Some useful and practical features of PoV models include estimation of interactions between latent variables, latent variable moderators, manifest moderators with missing values, and manifest or latent squared terms. Expected means and covariances are analytically derived for a simple product of two variables and it is shown that the method reproduces previously published results for this special case. It is shown algebraically that using centered multiplicands results in an unidentified model, but if the multiplicands have non-zero means, the result is identified. The method has been implemented in OpenMx and Ωnyx and is applied in five extensive simulations.

Anchor Point Selection: Scale Alignment Based on an Inequality Criterion.

For detecting differential item functioning (DIF) between two or more groups of test takers in the Rasch model, their item parameters need to be placed on the same scale. Typically this is done by means of choosing a set of so-called anchor items based on statistical tests or heuristics. Here the authors suggest an alternative strategy: By means of an inequality criterion from economics, the Gini Index, the item parameters are shifted to an optimal position where the item parameter estimates of the groups best overlap. Several toy examples, extensive simulation studies, and two empirical application examples are presented to illustrate the properties of the Gini Index as an anchor point selection criterion and compare its properties to those of the criterion used in the alignment approach of Asparouhov and Muthén. In particular, the authors show that-in addition to the globally optimal position for the anchor point-the criterion plot contains valuable additional information and may help discover unaccounted DIF-inducing multidimensionality. They further provide mathematical results that enable an efficient sparse grid optimization and make it feasible to extend the approach, for example, to multiple group scenarios.

Empathy, cognitive functioning, and prosocial behavior in mentored children.

Assessing the effect mentors have on their mentees is methodologically challenging: most programs merely provide relatively short mentoring durations (typically in the range of 1 year), age ranges are usually rather small, and examining dyads with anything other than questionnaires has proven to be challenging in the past. Thus, although some excellent causal studies do exist, in general causal research is limited in the field and studies are opened up to social desirability. Using a controlled laboratory setting, the current study investigates the causal effect of a mentor's presence on the mentee's empathic accuracy, cognitive functioning, and prosocial behavior. The sample is characterized by a wide age range for mentees and long mentoring durations. Results support the hypothesis that mentees' performance is improved in all three domains when their mentor is present as compared to when their mentor is absent. Furthermore, mentoring duration was positively associated with the mentee's cognitive functioning when controlling for the mentee's age. The current findings extend our knowledge of the benefits of youth mentoring programs and demonstrate the necessity to include laboratory research when investigating mentoring dyads.

Within-person structures of daily cognitive performance differ from between-person structures of cognitive abilities.

Over a century of research on between-person differences has resulted in the consensus that human cognitive abilities are hierarchically organized, with a general factor, termed general intelligence or "g," uppermost. Surprisingly, it is unknown whether this body of evidence is informative about how cognition is structured within individuals. Using data from 101 young adults performing nine cognitive tasks on 100 occasions distributed over six months, we find that the structures of individuals' cognitive abilities vary among each other, and deviate greatly from the modal between-person structure. Working memory contributes the largest share of common variance to both between- and within-person structures, but the g factor is much less prominent within than between persons. We conclude that between-person structures of cognitive abilities cannot serve as a surrogate for within-person structures. To reveal the development and organization of human intelligence, individuals need to be studied over time.

Ergodic Subspace Analysis.

Properties of psychological variables at the mean or variance level can differ between persons and within persons across multiple time points. For example, cross-sectional findings between persons of different ages do not necessarily reflect the development of a single person over time. Recently, there has been an increased interest in the difference between covariance structures, expressed by covariance matrices, that evolve between persons and within a single person over multiple time points. If these structures are identical at the population level, the structure is called ergodic. However, recent data confirms that ergodicity is not generally given, particularly not for cognitive variables. For example, the g factor that is dominant for cognitive abilities between persons seems to explain far less variance when concentrating on a single person's data. However, other subdimensions of cognitive abilities seem to appear both between and within persons; that is, there seems to be a lower-dimensional subspace of cognitive abilities in which cognitive abilities are in fact ergodic. In this article, we present ergodic subspace analysis (ESA), a mathematical method to identify, for a given set of variables, which subspace is most important within persons, which is most important between person, and which is ergodic. Similar to the common spatial patterns method, the ESA method first whitens a joint distribution from both the between and the within variance structure and then performs a principle component analysis (PCA) on the between distribution, which then automatically acts as an inverse PCA on the within distribution. The difference of the eigenvalues allows a separation of the rotated dimensions into the three subspaces corresponding to within, between, and ergodic substructures. We apply the method to simulated data and to data from the COGITO study to exemplify its usage.

Bayesian power equivalence in latent growth curve models.

Longitudinal studies are the gold standard for research on time-dependent phenomena in the social sciences. However, they often entail high costs due to multiple measurement occasions and a long overall study duration. It is therefore useful to optimize these design factors while maintaining a high informativeness of the design. Von Oertzen and Brandmaier (2013,Psychology and Aging, 28, 414) applied power equivalence to show that Latent Growth Curve Models (LGCMs) with different design factors can have the same power for likelihood-ratio tests on the latent structure. In this paper, we show that the notion of power equivalence can be extended to Bayesian hypothesis tests of the latent structure constants. Specifically, we show that the results of a Bayes factor design analysis (BFDA; Schönbrodt & Wagenmakers (2018,Psychonomic Bulletin and Review, 25, 128) of two power equivalent LGCMs are equivalent. This will be useful for researchers who aim to plan for compelling evidence instead of frequentist power and provides a contribution towards more efficient procedures for BFDA.

Lifespan Changes in Network Structure and Network Topology Dynamics During Rest and Auditory Oddball Performance.

Behavioral and physiological evidence suggests that developmental changes lead to enhanced cortical differentiation and integration through maturation and learning, and that senescent changes during aging result in dedifferentiation and reduced cortical specialization of neural cell assemblies. We used electroencephalographic (EEG) recordings to evaluate network structure and network topology dynamics during rest with eyes closed and open, and during auditory oddball task across the lifespan. For this evaluation, we constructed a hyper-frequency network (HFN) based on within- and cross-frequency coupling (WFC and CFC, respectively) at 10 oscillation frequencies ranging between 2 and 20 Hz. We found that WFC increased monotonously across the lifespan, whereas CFC showed a U-shaped relationship. These changes in WFC and CFC strengths coevolve with changes in network structure and network topology dynamics, namely the magnitude of graph-theoretical topology measures increased linearly with age (except for characteristic path length, which is going shorter), while their standard deviation showed an inverse U-shaped relationship with a peak in young adults. Temporal as well as structural or nodal similarity of network topology (with some exceptions) seems to coincide with variability changes, i.e., stronger variability is related to higher similarity between consecutive time windows or nodes. Furthermore, network complexity measures showed different lifespan-related patterns, which depended on the balance of WFC and CFC strengths. Both variability and complexity of HFNs were strongly related to the perceptual speed scores. Finally, investigation of the modular organization of the networks revealed higher number of modules and stronger similarity of community structures across time in young adults as compared with children and older adults. We conclude that network variability and complexity measures reflect temporal and structural topology changes in the functional organization and reorganization of neuronal cell assemblies across the lifespan.

Using the Dirichlet process to form clusters of people's concerns in the context of future party identification.

Connections between interindividual differences and people's behavior has been widely researched in various contexts, often by using top-down group comparisons to explain interindividual differences. In contrast, in this study, we apply a bottom-up approach in which we identify meaningful clusters in people's concerns about various areas of life (e.g., their own health, their financial situation, the environment). We apply a novel method, Dirichlet clustering, to large-scale longitudinal data from the German Socioeconomic Panel Study (SOEP) to investigate whether concerns of people living in Germany evaluated in 2010 (t0) cluster participants into robust and separable groups, and whether these groups vary regarding their party identification in 2017 (t0 + 7). Clustering results suggest a range of different groups with specific concern patterns. Some of these notably specific patterns of concerns indicate links to party identification. In particular, some patterns show an increased identification with smaller parties as the 'Bündnis 90/Die Grünen' ('Greens'), the left wing party 'Die Linke' ('The Left') or the right-wing party 'Alternative für Deutschland' ('Alternative for Germany', AfD). Considering that we identify as many as 37 clusters in total, among them at least six with clearly different party identification, it can also be concluded that the complexity of political concerns may be larger than has been assumed before.

Relational mobility predicts social behaviors in 39 countries and is tied to historical farming and threat.

Biologists and social scientists have long tried to understand why some societies have more fluid and open interpersonal relationships and how those differences influence culture. This study measures relational mobility, a socioecological variable quantifying voluntary (high relational mobility) vs. fixed (low relational mobility) interpersonal relationships. We measure relational mobility in 39 societies and test whether it predicts social behavior. People in societies with higher relational mobility report more proactive interpersonal behaviors (e.g., self-disclosure and social support) and psychological tendencies that help them build and retain relationships (e.g., general trust, intimacy, self-esteem). Finally, we explore ecological factors that could explain relational mobility differences across societies. Relational mobility was lower in societies that practiced settled, interdependent subsistence styles, such as rice farming, and in societies that had stronger ecological and historical threats.

Precision, Reliability, and Effect Size of Slope Variance in Latent Growth Curve Models: Implications for Statistical Power Analysis.

Latent Growth Curve Models (LGCM) have become a standard technique to model change over time. Prediction and explanation of inter-individual differences in change are major goals in lifespan research. The major determinants of statistical power to detect individual differences in change are the magnitude of true inter-individual differences in linear change (LGCM slope variance), design precision, alpha level, and sample size. Here, we show that design precision can be expressed as the inverse of effective error. Effective error is determined by instrument reliability and the temporal arrangement of measurement occasions. However, it also depends on another central LGCM component, the variance of the latent intercept and its covariance with the latent slope. We derive a new reliability index for LGCM slope variance-effective curve reliability (ECR)-by scaling slope variance against effective error. ECR is interpretable as a standardized effect size index. We demonstrate how effective error, ECR, and statistical power for a likelihood ratio test of zero slope variance formally relate to each other and how they function as indices of statistical power. We also provide a computational approach to derive ECR for arbitrary intercept-slope covariance. With practical use cases, we argue for the complementary utility of the proposed indices of a study's sensitivity to detect slope variance when making a priori longitudinal design decisions or communicating study designs.

Hyper-Transcranial Alternating Current Stimulation: Experimental Manipulation of Inter-Brain Synchrony.

We walk together, we watch together, we win together: Interpersonally coordinated actions are omnipresent in everyday life, yet the associated neural mechanisms are not well understood. Available evidence suggests that the synchronization of oscillatory activity across brains may provide a mechanism for the temporal alignment of actions between two or more individuals. In an attempt to provide a direct test of this hypothesis, we applied transcranial alternating current stimulation simultaneously to two individuals (hyper-tACS) who were asked to drum in synchrony at a set pace. Thirty-eight female-female dyads performed the dyadic drumming in the course of 3 weeks under three different hyper-tACS stimulation conditions: same-phase-same-frequency; different-phase-different-frequency; sham. Based on available evidence and theoretical considerations, stimulation was applied over right frontal and parietal sites in the theta frequency range. We predicted that same-phase-same-frequency stimulation would improve interpersonal action coordination, expressed as the degree of synchrony in dyadic drumming, relative to the other two conditions. Contrary to expectations, both the same-phase-same-frequency and the different-phase-different-frequency conditions were associated with greater dyadic drumming asynchrony relative to the sham condition. No influence of hyper-tACS on behavioral performance was seen when participants were asked to drum separately in synchrony to a metronome. Individual and dyad preferred drumming tempo was also unaffected by hyper-tACS. We discuss limitations of the present version of the hyper-tACS paradigm, and suggest avenues for future research.

Many-level multilevel structural equation modeling: An efficient evaluation strategy.

Structural equation models are increasingly used for clustered or multilevel data in cases where mixed regression is too inflexible. However, when there are many levels of nesting, these models can become difficult to estimate. We introduce a novel evaluation strategy, Rampart, that applies an orthogonal rotation to the parts of a model that conform to commonly met requirements. This rotation dramatically simplifies fit evaluation in a way that becomes more potent as the size of the data set increases. We validate and evaluate the implementation using a 3-level latent regression simulation study. Then we analyze data from a state-wide child behavioral health measure administered by the Oklahoma Department of Human Services. We demonstrate the efficiency of Rampart compared to other similar software using a latent factor model with a 5-level decomposition of latent variance. Rampart is implemented in OpenMx, a free and open source software.

Structure and Topology Dynamics of Hyper-Frequency Networks during Rest and Auditory Oddball Performance.

Resting-state and task-related recordings are characterized by oscillatory brain activity and widely distributed networks of synchronized oscillatory circuits. Electroencephalographic recordings (EEG) were used to assess network structure and network dynamics during resting state with eyes open and closed, and auditory oddball performance through phase synchronization between EEG channels. For this assessment, we constructed a hyper-frequency network (HFN) based on within- and cross-frequency coupling (WFC and CFC, respectively) at 10 oscillation frequencies ranging between 2 and 20 Hz. We found that CFC generally differentiates between task conditions better than WFC. CFC was the highest during resting state with eyes open. Using a graph-theoretical approach (GTA), we found that HFNs possess small-world network (SWN) topology with a slight tendency to random network characteristics. Moreover, analysis of the temporal fluctuations of HFNs revealed specific network topology dynamics (NTD), i.e., temporal changes of different graph-theoretical measures such as strength, clustering coefficient, characteristic path length (CPL), local, and global efficiency determined for HFNs at different time windows. The different topology metrics showed significant differences between conditions in the mean and standard deviation of these metrics both across time and nodes. In addition, using an artificial neural network approach, we found stimulus-related dynamics that varied across the different network topology metrics. We conclude that functional connectivity dynamics (FCD), or NTD, which was found using the HFN approach during rest and stimulus processing, reflects temporal and topological changes in the functional organization and reorganization of neuronal cell assemblies.

Handling Missing Data With Multilevel Structural Equation Modeling and Full Information Maximum Likelihood Techniques.

With increasing access to population-based data and electronic health records for secondary analysis, missing data are common. In the social and behavioral sciences, missing data frequently are handled with multiple imputation methods or full information maximum likelihood (FIML) techniques, but healthcare researchers have not embraced these methodologies to the same extent and more often use either traditional imputation techniques or complete case analysis, which can compromise power and introduce unintended bias. This article is a review of options for handling missing data, concluding with a case study demonstrating the utility of multilevel structural equation modeling using full information maximum likelihood (MSEM with FIML) to handle large amounts of missing data. MSEM with FIML is a parsimonious and hypothesis-driven strategy to cope with large amounts of missing data without compromising power or introducing bias. This technique is relevant for nurse researchers faced with ever-increasing amounts of electronic data and decreasing research budgets. © 2016 Wiley Periodicals, Inc.

Maintained Individual Data Distributed Likelihood Estimation (MIDDLE).

Maintained Individual Data Distributed Likelihood Estimation (MIDDLE) is a novel paradigm for research in the behavioral, social, and health sciences. The MIDDLE approach is based on the seemingly impossible idea that data can be privately maintained by participants and never revealed to researchers, while still enabling statistical models to be fit and scientific hypotheses tested. MIDDLE rests on the assumption that participant data should belong to, be controlled by, and remain in the possession of the participants themselves. Distributed likelihood estimation refers to fitting statistical models by sending an objective function and vector of parameters to each participant's personal device (e.g., smartphone, tablet, computer), where the likelihood of that individual's data is calculated locally. Only the likelihood value is returned to the central optimizer. The optimizer aggregates likelihood values from responding participants and chooses new vectors of parameters until the model converges. A MIDDLE study provides significantly greater privacy for participants, automatic management of opt-in and opt-out consent, lower cost for the researcher and funding institute, and faster determination of results. Furthermore, if a participant opts into several studies simultaneously and opts into data sharing, these studies automatically have access to individual-level longitudinal data linked across all studies.

Using within-subject pattern classification to understand lifespan age differences in oscillatory mechanisms of working memory selection and maintenance.

In lifespan studies, large within-group heterogeneity with regard to behavioral and neuronal data is observed. This casts doubt on the validity of group-statistics-based approaches to understand age-related changes on cognitive and neural levels. Recent progress in brain-computer interface research demonstrates the potential of machine learning techniques to derive reliable person-specific models, representing brain behavior mappings. The present study now proposes a supervised learning approach to derive person-specific models for the identification and quantification of interindividual differences in oscillatory EEG responses related to working memory selection and maintenance mechanisms in a heterogeneous lifespan sample. EEG data were used to discriminate different levels of working memory load and the focus of visual attention. We demonstrate that our approach leads to person-specific models with better discrimination performance compared to classical person-nonspecific models. We show how these models can be interpreted both on an individual as well as on a group level. One of the key findings is that, with regard to the time dimension, the between-person variance of the obtained person-specific models is smaller in older than in younger adults. This is contrary to what we expected because of increased behavioral and neuronal heterogeneity in older adults.

LIFESPAN: A tool for the computer-aided design of longitudinal studies.

Researchers planning a longitudinal study typically search, more or less informally, a multivariate space of possible study designs that include dimensions such as the hypothesized true variance in change, indicator reliability, the number and spacing of measurement occasions, total study time, and sample size. The main search goal is to select a research design that best addresses the guiding questions and hypotheses of the planned study while heeding applicable external conditions and constraints, including time, money, feasibility, and ethical considerations. Because longitudinal study selection ultimately requires optimization under constraints, it is amenable to the general operating principles of optimization in computer-aided design. Based on power equivalence theory (MacCallum et al., 2010; von Oertzen, 2010), we propose a computational framework to promote more systematic searches within the study design space. Starting with an initial design, the proposed framework generates a set of alternative models with equal statistical power to detect hypothesized effects, and delineates trade-off relations among relevant parameters, such as total study time and the number of measurement occasions. We present LIFESPAN (Longitudinal Interactive Front End Study Planner), which implements this framework. LIFESPAN boosts the efficiency, breadth, and precision of the search for optimal longitudinal designs. Its initial version, which is freely available at http://www.brandmaier.de/lifespan, is geared toward the power to detect variance in change as specified in a linear latent growth curve model.

Efficient Hessian computation using sparse matrix derivatives in RAM notation.

This article proposes a new, more efficient method to compute the minus two log likelihood, its gradient, and the Hessian for structural equation models (SEMs) in reticular action model (RAM) notation. The method exploits the beneficial aspect of RAM notation that the matrix derivatives used in RAM are sparse. For an SEM with K variables, P parameters, and P' entries in the symmetrical or asymmetrical matrix of the RAM notation filled with parameters, the asymptotical run time of the algorithm is O(P ' K (2) + P (2) K (2) + K (3)). The naive implementation and numerical implementations are both O(P (2) K (3)), so that for typical applications of SEM, the proposed algorithm is asymptotically K times faster than the best previously known algorithm. A simulation comparison with a numerical algorithm shows that the asymptotical efficiency is transferred to an applied computational advantage that is crucial for the application of maximum likelihood estimation, even in small, but especially in moderate or large, SEMs.

Optimal study design with identical power: an application of power equivalence to latent growth curve models.

Structural equation models have become a broadly applied data-analytic framework. Among them, latent growth curve models have become a standard method in longitudinal research. However, researchers often rely solely on rules of thumb about statistical power in their study designs. The theory of power equivalence provides an analytical answer to the question of how design factors, for example, the number of observed indicators and the number of time points assessed in repeated measures, trade off against each other while holding the power for likelihood-ratio tests on the latent structure constant. In this article, we present applications of power-equivalent transformations on a model with data from a previously published study on cognitive aging, and highlight consequences of participant attrition on power.